Intelligent indexes with inheritable fuzzy relations

Abstract

The paper outlines an approach to building "intelligent" indexes for digital libraries. The index takes the form of a concept taxonomy, wherein concepts represent classes of objects, and arcs are restricted to being either: (i) parent-child specificity links, where child classes may be either fully or partially contained within their parents; or (ii) is-related-to links, where no particular subclass/superclass interpretation is implied. The taxonomy is provided an underlying layer of logical intelligence that makes it function as an expert system that can be queried by the user as to the deeper logical interrelations between the concepts in the taxonomy, as well as the objects thus classified. The system applies a new solution to the problem of default reasoning with exceptions in multiple inheritance reasoning systems. In effect, is-related-to links are inheritable downwards in the taxonomy, unless this inheritance is blocked by counteracting information.